Research on X-ray coronary artery branches instance segmentation and matching task.

Area of Science:

• Medical Imaging
• Computer Vision
• Biomedical Engineering

Background:

• 3D reconstruction of coronary arteries from X-ray images presents challenges in matching vessel branches across different viewpoints.
• Distinguishing between left and right coronary artery branches complicates the segmentation and matching process.

Purpose of the Study:

• To develop an accurate method for 3D coronary artery reconstruction by addressing the challenge of vessel branch matching.
• To propose a specialized instance segmentation network (YOLO-CAVBIS) for handling deformed and dynamic coronary vessels.

Main Methods:

• A coronary artery classification dataset was created to distinguish left and right coronary arteries using YOLOv8-cls.
• Classified images were processed by two parallel YOLO-CAVBIS networks for instance segmentation of coronary artery branches.
• Matching of vessel branches across different viewpoints was performed based on color consistency.

Main Results:

• The coronary artery classification model achieved 100% accuracy.
• The YOLO-CAVBIS model demonstrated high performance with mAP50 scores of 98.4% for left coronary branches and 99.4% for right coronary branches.
• The YOLO-CAVBIS network showed superiority in extracting features from deformed and dynamic vessels compared to other networks.

Conclusions:

• The proposed YOLO-CAVBIS network is effective for coronary artery branches instance segmentation, particularly for deformed and dynamic vessels.
• This method provides a robust baseline for 3D coronary artery reconstruction tasks.
• The approach significantly enhances the accuracy and specificity of vessel branch matching in medical imaging.